1 of 52

Confrey Alianji

LinkedIn: https://www.linkedin.com/in/confreymungaualianji/

Confrey is an innovation expert and venture builder with experience as an HCD Facilitator. Confrey has dedicated his experience to exploring how human-centred design (HCD) can improve sexual and reproductive health needs and rights for youth and adolescents. He is part of the community of practice that drives, shares, and increases learning to help shape the SRHR field. Working with AgaKhan Foundation-AKF, Population Service International, among others. He brings a lot of experience in program design, managing 80% of the Innovation Enabler Africa Region, responsible for over 17 countries at WWF International. He is also responsible for starting up the WWF-Kenya Innovation Program, “PandaLabs''. He also coordinates WWF's Finance Practice Startups and Ventures Workstreams, working with startups in Asia and Africa. He is also a business and strategy designer who serves the innovation needs of various impact organizations, African entrepreneurs, and businesses. He has over ten years of working experience in the region.

Consultant 1

��WWF UK – Intro to Artificial Intelligence�FEBRUARY 2024

2 of 52

AGENDA

Intro to AI – 15mins

Why do we need to care about AI now?
What is AI and how does it work?
Different types of AI
How might we integrate AI into our work

Demo - How you can start using CoPilot now – 15mins

AI in action at the WWF - 10mins

Q&A

Over the next 60 minutes we are going to cover a broad introduction and give you some practical ways for how you can use AI today in your work

The first part will cover what AI is, how it works and how we might integrate it into our work, and will take about 20mins
The second part will be where James takes you through some practical ways you can use CoPilot in your roles today and you’ll be able to follow along yourself if you like
And lastly, but really excitingly, we will be joined by Rod who is going to talk you through how we are already using AI in some of our work in the Artic

And at the end we will have time for a Q&A. As we are going through this session, please drop questions in the chat. We will try to answer them all at the end and if we don’t have time we will share a Q&A response afterwards

3 of 52

WE ARE NOT EXPERTS

(But we are enthusiastic and interested in AI)

4 of 52

WHY DO WE NEED TO CARE ABOUT AI NOW?

5 of 52

History of generational tech shifts

There has been a history of generational shifts, that have impacted how we live and work (CLICK)

The electrification of the world, transformed manufacturing and society in the industrial revolution and brought in the Technological Revolution (CLICK)

The Internet gave way to the Information and Digital Revolution with the launch of the world wide web in 1994. Nobody at that time foresaw how the way in which we shop would transform, and it took nearly a decade for online shopping to become mainstream and for us to see the economic impact on the high street, and how it had to adapt. (CLICK)

Social media transformed how we communicate and connect with each other in 2004, but again we are only now beginning to really see the impacts social media is having on society.

Now these shifts were slower burners, giving us time to adapt and gradually adjust and plan for and they tended to come along every decade or so. (CLICK)

We have had 2 of these in generational shifts in the last 3 years. COVID forced workforces to go remote and now we live in a hybrid world. We adapted immediately and just two years on employers and employees are figuring out what works best. (CLICK)

And then there is AI. How we decide to integrate AI in our work will transform how we work and how we can deliver our mission. AI system capabilities will double every 6 months, this isn’t going to be a slow decade long transformation, we will start to see change and impacts within a year.

6 of 52

THE FUTURE OF HOW WE USE AI IS NOT A PREDETERMINED DESTINY

7 of 52

UTOPIAN

DYSTOPIAN

Created by DALLE: MS Co Pilot

When any new technology comes online, or major change happens, you will tend to fall somewhere on this spectrum of Utopian to Dystopian

Consider remote working. On the Utopian side people thought it would give great work life balance, increase productivity, give employees empowerment.

And on the Dystopian side that it would reduce productivity, be the death of the high street and to a collapse of the economy and peoples sense of self. In reality we landed somewhere in the middle.

Contemporary discourses of AI range from dramatically increasing productivity to leading to human extinction, and this reflects the same sociotechnical dynamics we have seen in the past. Whether it was the introduction of personal computers or remote working.

If we look at our survey results for AI we have nearly 50% of people feeling positive, but we have 35% neutral and 15% negative. So, we have a real spectrum of views towards AI, and that is completely normal. And if you are one of those people who are more on the dystopian side, do not fear, those feelings of worry or negativity about AI are completely natural. I would say 12 months ago I sat in that dystopian viewpoint, and over the past 12 months I have gradually made my way more to the middle.

Now the reason I bring this up is because the context to which a new technology lands and how it is interpretated by people, determines how it is adopted and with what consequences.

If everyone had a dystopian view and felt really scared about AI, it would never be adopted, so it is crucial that we ensure everyone at the WWF is educated on and aware of all the opportunities, as well as the challenges, that AI brings, and trained on how to utilise it, so we feel confident when we decide to integrate it into our work.

And this is extremely important now, as AI is already here and developing quickly.

8 of 52

How might we integrate AI?

Human-centered approach

Before we jump into look at what AI is and get into some practical training examples. It’s important to think about how we might integrate AI into our work.

Like we said, our the future of how we work is not a predetermined destiny, we get to shape it.

And we have seen before how technology can be integrated into work, varying across work domains and tasks and this image I think is a good way to view the different ways we could integrate AI into our work

L represents the load/task to be completed

H represents the human

C represents computer/technology

(CLICK) Our first option is that we complete the task on our own (CLICK)
Or we can get computers to share the task with us by extending our own human capabilities or relieving us to make the job easier, (An example here is using MS Word rather than writing everything out on pen and paper) (CLICK)
Or we can get computers to trade loads with us by being a back-up in case the we falter, or completely replacing us where tasks can be fulfilled by new technology (Spell Check is there to back you up. Car manufacturing or bomb disposal is now done by machines for efficiency and quality reasons and safety reasons) (CLICK)

This human-centered approach to AI takes a complementary perspective, in which humans and AI are partners that balance out each other’s weaknesses. And that is the approach we want to take at the WWF and I’d encourage you as we are going through this training to think about where AI might be able to share the load and where it might be able to trade tasks with you

9 of 52

WHAT IS AI?

10 of 52

WHAT IS AI?

AI refers to the development of computer systems that can perform tasks that typically require human intelligence. These tasks include understanding natural language, recognizing patterns, making decisions, and learning from data.

AI is not synonymous with human-like consciousness; it's a tool for processing and interpreting information.

AI is really tricky to define for a multitude of reasons, and even AI researchers don’t have an agreed definition. But at the WWF we have decided to use this simplistic definition to serve as an overarching definition.

AI is sometimes thought of a system in itself, but it’s actually a set of technologies implemented into a system to let it reason, learn and solve a complex problem. And we’ll cover more of the elements that come together for AI to work later on

The keys things to pull out here to help you understand AI is the process of – understanding, recognising patterns, learning from the data and making predictions. What makes AI so powerful is that this process can be a circular cycle, where the systems can continue to learn and develop

The last point on this slide is really important…AI does not have human-like consciousness….

11 of 52

Narrow AI	General AI
Trained and focused to perform specific tasks. Narrow AI drives most of the AI that surrounds us today. E.g., Apple Siri, Amazons Alexa, Autonomous Vehicles.	Theoretical form of AI where a machine would have an intelligence equal to humans; it would have a self-aware consciousness that has the ability to solve problems, learn, and plan for the future. Entirely theoretical with no practical examples in use today, that doesn't mean AI researchers aren't also exploring its development.

vs

…and when defining and talking about AI we have to be cautious as many of the words that we use can be quite misleading. For example ‘intelligence’

You may well say, that a system is intelligent, perhaps because it delivers accurate navigation instructions or detects signs of cancer in photographs of skin lesions. When we hear something like this, the word “intelligent” easily suggests that the system is capable of performing any task an intelligent person is able to perform, going to the supermarket and cooking dinner, washing and folding laundry, and so on. This can be misleading because artificial intelligence is narrow, being able to solve one problem tells us nothing about the ability to solve another, different problem.

Narrow AI is AI trained and focused to perform specific tasks and it drives most of the AI that surrounds us today. E.g. Apple Siri can perform specific tasks such as setting reminders, answering questions, and navigating through applications, Autonomous Vehicles can identify hazards and drive us from A-B.

General AI is a theoretical form of AI where a machine would have an intelligence equal to humans; it would have a self-aware consciousness that has the ability to solve problems, learn, and plan for the future across a wide range of tasks….this is what you have probably seen in films and TV shows, where their abilities mirror human cognitive abilities. While General AI is still entirely theoretical with no practical examples in use today, that doesn't mean AI researchers aren't exploring its development.

12 of 52

WHAT IS AI?

A

B

C

A = Computer Science is the development of computer systems

B = AI is the development of the intelligence of machines or software

C = Machine learning is considered part of AI, it involves developing algorithms that enable computers to learn from and make predictions or decisions based on data.

To understand what AI is and what technologies are driving the AI around us, it helps to understand the organisation of AI within wider computer science discipline

A – Computer Science focuses on methods involved in designing and programming of human-made computing systems. (CLICK) And AI sits within this

B – AI (Artificial Intelligence) - is the intelligence of machines or software, as opposed to the intelligence of humans or animals. It is also the field of study in computer science that develops and studies intelligent machines. (CLICK)

C – Machine Learning – considered part of AI – involves developing algorithms that enable computers to learn from and make predictions or decisions based on data.

As we have mentioned AI systems can be trained to perform specific tasks. Now while we don’t need to know the detailed technical information, it’s good to have a basic understanding of how AI works, and the secret lies in a machine learning. (CLICK)

13 of 52

Data & Information

Machine learning algorithms learn to understand the data, recognise patterns and make predictions

The models are the output of what is learnt from the algorithms

Source: Google, Qwiklabs Course – Intro to AI

HOW AI WORKS

INPUT

OUTPUT

14 of 52

Curation AI	Generative AI
Is the use of algorithms to process huge volumes of data, deciphering meaning, and patterns. E.g., Instagram or Netflix AI algorithm works by analysing user data to make sense of user intent, thereby helping marketers address consumers better.	Is capable of generating text, images, or other media, using generative models. Generative AI models learn the patterns and structure of their input training data and then generate new data that has similar characteristics e.g., Chat GPT, or Adobe Firefly can generate new text and images based on your text prompts

vs

Machine learning has been around for decades and has driven most of the AI we have come into contact with up to now, which is Curation AI.

Curation AI is the use of ML algorithms to process huge volumes of data, and decipher meaning and patterns. Social media was the first mass human contact with AI. Social media companies use Curation AI systems to process huge amounts of behavioural data on you, to try and predict what content they need to show you next to keep your attention. Netflix analyses all your viewing habits and suggests the next thing for you to watch to keep you on Netflix for longer.

This second contact with AI which we are experiencing now is known as ‘Creation AI’, AI which helps us create ‘things’ or solve problems. It is capable of generating text, images, or other media. Generative AI learn the patterns and structure of their input training data and then generates new data that has similar characteristics.

E.g. ChatGPT and CoPilot uses all public available information and data from the internet to learn from. Bairbie.me was an AI which was trained on what a Ken Doll looks like, so what are the patterns within 1000s of ken doll images that make it a ken doll. It is then able to take my photo, analyses it and using its model, turn my image into Ken form.

But how is Generative AI able to do this? (CLICK)

15 of 52

WHAT IS GENERATIVE AI?

A

B

C

D = Deep Learning – this is a subset of Machine Learning (ML).

It uses Artificial Neural Networks – allowing them to process more complex patterns than traditional machine learning.

D

16 of 52

WHAT IS GENERATIVE AI?

Artificial Neural Network

Human Neural Network

…and are made up of many interconnected ‘virtual’ neurons or nodes.

Like the neurons in our brains, which build connections and patterns based on the stimulus we receive via our senses, these artificial neurons allow the AI systems to process more complex patterns than traditional machine learning.

The human brain has billions of neurons connected by synapses, and similarly, ANNs have a network of nodes connected by lines or pathways.

These networks are organized in layers: an input layer to receive the data (yellow), hidden layers to process the data (Orange), and an output layer to deliver the final result (Teal). Think of it as a complex system of filters where each layer learns to recognize different patterns and features of the data.

ANNs learn from data through a process called "training". During training, the network is fed large amounts of data (for example, pictures of cats and dogs). It then tries to make predictions or classifications based on this data. If it makes a mistake, the error is used to adjust the connections between the nodes (similar to strengthening or weakening synapses in the human brain), gradually improving its accuracy.

17 of 52

WHAT IS GENERATIVE AI?

A

B

C

D

Generative AI – this is also a subset of Deep Learning

18 of 52

HOW DO GENERTIVE AI TOOLS WORK?

19 of 52

User Prompt sent to CoPilot

CoPilot sends prompt to its LLM

Co Pilot receives LLM response

Co Pilot sends user response

1

2

3

4

Using everything, we have learnt, this is how CoPilot works (CLICK)

You send copilot a prompt – a prompt is an instruction. The better the prompt the better the quality of the response. – “Write me a poem about AI’ (CLICK)
That prompt is sent to the language model which is trained to understand all data and information publicly available on the internet using deep learning and it understands human language, and as a result can respond using human language. (CLICK)
The model then uses what it has learnt about poetry and creates a response, which is then sent to copilot (CLICK)
CoPilot then responds to the users prompt with a poem about AI. And this all happens in less than 30s
Then if the user wants, they can ask further prompts to refine their poem, maybe they want it so it’s suitable for a 5 year old or they want it in French, and the process repeats.

CoPilot has learned from a huge amount of text information, taken from the internet. It doesn't really understand what it reads, but it remembers how words and sentences fit together in patterns. But sometimes, it might guess wrong because after all, it's only just guessing, based on what it has seen before, not because it truly understands what it's saying.

20 of 52

Models trained on vast amounts of data

= LARGE LANGUAGE MODELS (LLM)

21 of 52

HOW WE MIGHT INEGRATE AI

Write email and send myself

AI writes and sends

I write and send email, but AI support is there if I need help crafting, or I forget to send the email

I write and AI automatically corrects spelling and grammar issues and suggests tone changes

AI writes the draft. I then edit before sending

So now we understand in really basic terms how AI works we can begin to think about how it might work with us on some of our work tasks.

Going back to our earlier model of how humans and technology integrate, here is a really simple example of how we might choose to integrate AI

Using the simple task of sending an email to a colleague. Imagine MS have a Copilot for Outlook, which they are working on. Think of it as our own AI assistant available to use on emails (CLICK)

We might decide that for emails we’re just going to write and send them manually ourselves as we have been doing (CLICK)
Or to make it easier for us, we’re going to use AI for spell check, predictive text and suggestions for changes to the tone and structure once I’ve written the email. So I’m doing the bulk of the task and AI is helping optimise it (CLICK)
We might decide that we want to use AI to write the first draft of the email, we might give it a quick prompt and off it goes and writes the email I want. Meaning rather than creator I become the editor before sending it, which then saves me time (CLICK)
I could also use the AI as a back up, so if I struggle writing an email, I can ask for help. Or if have a meeting coming up[ and I haven’t sent an agenda it might give me a reminder to send the agenda (CLICK)
Or the next stage which is full automation, the AI write and sends the email. It knows I’ve had a meeting with James on AI, and there is an action for me to send this deck to James. The AI will send that automatically.

In this example you can quickly see how AI can be a really supportive assistant and improve someone's output. You can also see how someone with learning difficulties or accessibility needs could really benefit from AI in supporting them with a regular day to day task such as writing an email.

As James goes through some use cases, start thinking about tasks you do in your role, and the extent to which you could collaborate with AI.

22 of 52

AI USE CASES

23 of 52

24 of 52

MICROSOFT COPILOT DEMO

https://copilot.microsoft.com/

Chats in Microsoft Copilot are secure. User and business data is protected and will not leak outside the organization. You can be confident that chat data is not saved, Microsoft has no eyes-on access to it, and it is not used to train the models.

25 of 52

DEMO 1 – Tell Me

26 of 52

DEMO 2 – Rephrase

27 of 52

DEMO 3 – Act Like

28 of 52

MICROSOFT COPILOT EXAMPLES

Persona	Tell the tool what its job title is (role)
Objective	What do you want to do?
Audience	Who will receive the message?
Output Parameters	Tone, style, length, let the tool know any guidance
Context	What points should be covered? What's the call to action?

29 of 52

MICROSOFT COPILOT EXAMPLES

Persona	Act as if you are the best local tour guide for Copenhagen
Objective	Design me a 2-day itinerary for a wonderful experience
Audience	We are a married couple who have never been to Copenhagen and will be visiting in January. We're looking for experience true local Copenhagen and hidden gems
Output Parameters	Summarise the outputs in a table by day, and segment by morning, afternoon, evening and nighttime
Context	We are keen walkers and would love recommendations for walks in Copenhagen. We love doing a walking tour when we arrive to explore the city, as well as visiting museums and sights. Please include recommendations for authentic and affordable food and drink experiences for breakfast, coffee, lunch, dinner, and evening drinks.

DESK RESEARCH

30 of 52

IceNet AI Sea Ice Model and a Caribou feasibility study

31 of 52

Observations

Daily sea ice concentration forecast

IceNet Sea Ice Model (AI deep learning)

32 of 52

Multiple satellite image data sources

Moderate Resolution Imaging Spectroradiometer (MODIS)

250 m resolution
Daily return period
Limitation- no data in cloudy/ poor lighting conditions

Synthetic Aperture Radar (SAR)

40 - 80 m resolution
~ weekly return period
Limitation- Low return period, need to mosaic images

Ocean and Sea Ice

Satellite Application Facility (OSI-SAF)

25 km resolution
Daily return period
Limitation- Coarse grid cell size

Advanced Microwave Scanning Radiometer (AMSR2)

6 km resolution
Daily return period
Limitation- imagery only captured since 2013.

33 of 52

Observations

Daily sea ice concentration forecast

IceNet Sea Ice Model (AI deep learning)

34 of 52

35 of 52

The IceNet-Caribou feasibility study

Background & Motivation

Declining Dolphin and Union (DU) caribou

Population: 18,000 (2015)→3,800 (2020)¹
Endangered (COSEWIC, 2017)²
Declining sea ice levels make DU caribou migration more perilous^3,4

Environment and Natural Resources, Government of Nunavut
Committee on the status of Endangered Wildlife in Canada (COSEWIC)
Ekaluktutiak Hunters and Trappers Organization, Ice Breaking Workshop Summary Report. (2019)

Fall migration (VI→mainland)

Spring migration (mainland→VI)

36 of 52

The IceNet-Caribou feasibility study

Goal

Explore IceNet to predict timing and location of caribou migration over the Coronation Gulf

Objectives

Use satellite observations of sea ice concentration (SIC) to establish links with caribou migration timing/location
Use the IceNet artificial intelligence (AI) forecasting system to predict when/where crossings likely
Build alert system to inform local conservation groups (e.g. Gov. of Nunavut)

37 of 52

The IceNet-Caribou feasibility study

Wikimedia Commons

38 of 52

Caribou tracking data

DU caribou migrate across gulf:

In spring before the ice melts, to access calving grounds on VI
In autumn wait for ice to reform before crossing to mainland

Collar data for 161 caribou
Collected in:

1987 - 1989
1996 - 2006
2015 - present

Photo credit: WWF Canada

39 of 52

Autumn migration animation

40 of 52

SIC-migration relationship

41 of 52

Example Forecast

Crossings made when SIC threshold of 90% (± 7.28 days)
Could be used to predict two week migration start window
Early warning system for conservation planning – e.g. interaction with shipping activity.

42 of 52

Highlights

Research shows potential of predicting migrations using sea ice forecasts
Novel algorithm for automatically detecting caribou migration starts
Proof of concept caribou migration alert system

Next steps & future application

Integrate IceNet-Caribou system & Notification to Mariners
Polar Bear/ Human conflict early warning system
Whale migration/ blue corridors

43 of 52

AI: Implications and Applications for WWF UK
WWF AI Usage Guidelines
WWF AI Teams Group

FIND OUT MORE…

Reading and Training Resources

We will follow up with a list of reading and training resources.

This will keep building and there will be an AI learning pathway and resource hub later this year

44 of 52

Question Time��For any questions we don’t get around to answering, we’ll share answers after along with slides

45 of 52

APPENDIX

46 of 52

MICROSOFT COPILOT EXAMPLES

Persona	Act as an expert copywriter
Objective	Create a summary article which explains the Living Planet report in a way a 7 year old would understand
Audience	It will be read by children aged 7 years old
Output Parameters	Write a 200 word summary article on the finding of the Living Planet report
Context	The article should be warm and friendly in tone and ensure complicated elements are simplified. The article should focus on the main findings and causes of the issues we face as well as what we need to change to make things right

ADAPTING CONTENT TONE

47 of 52

MICROSOFT COPILOT EXAMPLES

Persona	Act as if you are a BBC Editor
Objective	Create an article on the impact of the labour party entering government in 2024 and the impact on the UK's environment and bio diversity
Audience	WWF workforce who will have a mixture of political knowledge, so please ensure the paper is simple, clear and concise.
Output Parameters	Please write 200 words max. Include a eye catching headline. Section headings should be bold and underlined. Please generate me an image to accompany this article that encapsulates the points raised within the article
Context	The article structure should include an overview of the key challenges and potential changes we might see from Labour, the impact on different sectors across the UK with particular focus on green and environmental issues, and a summary of the impact individuals might see.

DRAFT DOCUMENTS

48 of 52

HOW MIGHT WWF USE THESE TOOLS?

Productivity: Helping us work more efficiently. There are already tools offering automated meeting summaries, creation of documents & presentations.

Data Analysis & Coding: Copilot & Chat GPT can be helpful for various types of coding from python to Excel formulas. There are copilots available for tools like PowerBI, Excel & Github already.

Impact Measurement: Analysing satellite imagery for deforestation or counting species.

Content Creation: Creating drafts, mock-ups or quickly tailoring content to different audiences.

Engaging Supporters: AI can help us personalise supporter communications like enquiries & thank you letters at scale, improving engagement and retention rates.

Predictive Modelling: AI tools and machine learning can help build accurate models to identify potential donors or those likely to lapse and suggest optimal fundraising strategies.

Chatbots: custom LLMs can be created on specific sources on information (e.g. wwf.org.uk or HR policies) & answer questions based on the content.

There is a huge number of potential use cases within an organisation as large and diverse as WWF, below are a just a few examples.

49 of 52

STAYING SAFE WITH AI

It's crucial to utilise AI in ways that align with WWF policies and standards, including Data Protection, Information Security and Acceptable Use policies. Always review anything before putting it into use.

Remember we must never share any WWF supporter data or confidential information outside of the organisation.

A high-level guidance document has been created which is available on the Compliance Hub including 7 principles we should be aiming to follow:

1. Do no harm - AI systems should not be used in ways that cause or exacerbate harm, whether individual or collective

2. Fairness - AI systems should treat all people fairly without discrimination.

3. Reliability and safety - AI systems should perform reliably and safely.

4. Privacy and security - AI systems should be secure and respect privacy.

5. Inclusiveness - AI systems should empower everyone and engage people.

6. Transparency - AI systems should be understandable and clearly stated when used

7. Human Oversight - People should be accountable for AI systems and their outputs

50 of 52

COMING SOON - MICROSOFT COPILOTS FOR 365

Microsoft are rolling out a whole host of new Copilots across their products.

This is likely to present a way that WWF can trial some of these tools in a safe & managed way, ensuring we get value for the additional investment.

51 of 52

52 of 52

RAPID CAPABILITY IMPROVEMENT

Like we said, AI has been around for over 20 years, but the past 5 years have seen the most dramatic improvement in capabilities which is why we have seen this explosion in generative AI systems such as ChatGPT and why now it is being integrated widely across industries.

The capabilities of AI systems have been rapidly accelerating. This graph shows 0 as human performance benchmark.

In RED you can see Handwriting recognition took 15 years to get to a stage where it could perform better than a human
In GREEN Speech recognition took longer, Apple brought it in with Siri in 2010, which is able to recognise human language and then retrieve information from your phone initially, and then as it developed, from the web. Remember how bad it was in 2010, but it had rapid improvements and within 5 years was as good as a human
But look at how rapidly image recognition (Blue) and language understanding (Purple) improved – In PURPLE it took only a year for AI systems to understand language patterns and be able to generate new content.

What we have seen is capabilities slowly build for a decade and then accelerate in the last 5 years. Moore’s Law is the idea that technology capability will double every two years. This goes out the window , as with AI, as we mentioned earlier, capabilities are set to double every 6 months rather than two years.